Related Experiment Video
Updated: Aug 5, 2026
06:41
In Vivo Functional Study of Disease-associated Rare Human Variants Using Drosophila
Published on: August 20, 2019
LLM-Assisted Reanalysis of Unsolved Rare Disease Genomes Increases Diagnostic Yield
Aaron Jaech1, Morgan Cheatham2,3, Suyash S Shringarpure1
1OpenAI, San Francisco, CA, USA.
NEJM AI
|July 30, 2026
Summary
A large language model (LLM) assisted in reanalyzing rare genetic disorders, improving diagnostic yield by 4.8% in retrospective cases. This AI-driven approach identified new diagnoses and generated novel biological hypotheses for undiagnosed conditions.
Area of Science:
- Genomics and Bioinformatics
- Artificial Intelligence in Medicine
- Rare Disease Diagnostics
Background:
- Millions affected by rare and undiagnosed genetic disorders globally.
- Prolonged diagnostic odysseys due to limitations in conventional genomic interpretation.
- Need for repeated genomic analysis as knowledge evolves.
Purpose of the Study:
- To evaluate a large language model (LLM)-assisted workflow for retrospective reanalysis of rare genetic disorders.
- To assess the impact of LLM-driven interpretation on diagnostic yield and hypothesis generation.
Main Methods:
- Retrospective multicohort reanalysis using an LLM-assisted workflow.
- Ingestion of clinician notes, Human Phenotype Ontology (HPO) terms, and variant tables.
- Expert adjudication of candidate hypotheses based on established clinical guidelines.
Main Results:
- Overall diagnostic yield of 4.8% (18 of 376 cases) across four cohorts.
- Identification of seven previously unrecognized pathogenic findings ('rediscoveries').
- Generation of testable biological hypotheses, including a novel S1PR1 and vitiligo association.
Conclusions:
- LLM application in retrospective analysis significantly enhances diagnostic yield for rare diseases.
- The workflow effectively surfaces overlooked pathogenic findings and generates biologically grounded hypotheses.
- Prospective multicenter evaluation is warranted to validate these findings.
